learn

Learn

Explicitly teach the agent something to remember across all future sessions.

Usage: /learn [content] or /learn then describe what to save.

---

Step 1: Get the Content

If content was provided after the command, use it directly. If not, ask:

What do you want me to learn? Share a fact, preference, domain detail, or anything you want me to remember.

---

Step 2: Read vault.yml

If vault.yml exists, read it and extract folders.agent. Default to 05-agent if the file does not exist or the key is absent. Set agent_folder for all paths below.

---

Step 3: Classify the Input

Apply this rule to determine the destination:

→ [agent_folder]/context/ if the content describes the user's world:

• Domain, industry, or field knowledge
• Technical stack, tools, or systems
• Product, team, customers, or company
• Terminology, naming conventions, or abbreviations

→ [agent_folder]/memory/ if the content is a behavioral preference or pattern:

• How the user wants to be communicated with
• Things that help or hinder the AI's responses
• Recurring preferences or frustrations

Classification examples:

Input	Destination
"My stack is Go + Postgres + Redis"	context/
"When I say 'the app' I mean my SaaS at app.example.com"	context/
"My target user is small business owners, not enterprises"	context/
"I prefer async communication and document decisions in notes"	memory/
"I get frustrated when responses are too long"	memory/

Tiebreaker for domain-preference hybrids: If the input is a preference about how to work within a domain (e.g., "always use Go idioms, not Java-style abstractions" or "keep SQL queries readable, not optimized"), classify it as memory/ : it governs AI behavior, even though it references a technical domain. When in doubt: if you would change how you respond based on this input, it's memory/.

If classification is unclear, ask: "Is this about your world (context) or how you want me to behave (preference)?"

---

Step 4: Determine File Name

For context/ notes:

• File name: Topic Name.md (Title Case, by topic)
• Check if a relevant context note already exists in [agent_folder]/context/. If so, offer to append to it rather than create a new file.

For memory/ notes:

• File name: YYYY-MM-DD-slug.md where slug is a short kebab-case description (first note of the day)
• If a memory note already exists today (check [agent_folder]/memory/YYYY-MM-DD-*.md), use a counter: YYYY-MM-DD-02-slug.md, YYYY-MM-DD-03-slug.md, etc.
• Example (first note of day): 2026-03-23-prefers-async-comms.md
• Example (second note of day): 2026-03-23-02-no-long-responses.md

---

Step 5: Write the Note

For context/ notes:

---
tags: [agent-context]
created: YYYY-MM-DD
source: /learn
---

# [Topic]

[Content as provided, lightly structured if helpful]

When appending to an existing context note: Add a new ## section with a descriptive heading for the additional information. Do not rewrite or restructure existing content.

For memory/ notes:

---
tags: [agent-memory]
created: YYYY-MM-DD
source: /learn
---

# [Short description of pattern]

[Pattern or behavioral observation, 1-3 sentences]

---

Step 6: Confirm

Report what was saved:

Learned: "[short summary]" Saved to: [agent_folder]/[context or memory]/[filename]

I'll use this in future sessions when relevant.

If a context note was appended rather than created:

Updated: [agent_folder]/context/[filename] with new information.